Stanford’s SleepFM AI model can classify over 130 diseases from a single night of sleep data, up to six years before symptoms appear. Trained on 585,000 hours of recordings, it reframes sleep as one of the most information-rich diagnostic tools medicine has largely ignored.
What SleepFM Actually Found
Stanford’s model predicts systemic disease risk from one night of sleep recordings. The performance numbers are striking: mortality prediction reached a C-index of 0.84, dementia prediction 0.85, and myocardial infarction 0.81. Atrial fibrillation classification hit 0.81 AUC, nearly matching models built exclusively for that task.
What sets SleepFM apart is its generalist design. Earlier AI sleep tools targeted one condition at a time. This model classifies over 130 conditions from the same recording, functioning more like a broad health screening than a targeted test.
Limits and What Comes Next
SleepFM is a proof of concept, not a finished clinical product. It currently requires clinical-grade lab recordings costing $1,000 to $3,500, placing it out of reach for many. Training data diversity also needs broader validation before any clinical deployment.
If future versions can work with consumer wearables, even at reduced accuracy, the implications for population-scale screening become significant. The direction is clear: sleep is one of the most data-rich events your body produces, and science is finally learning to read it.
